WO2021207237A1 - Deep learning accelerator and random access memory with a camera interface - Google Patents
Deep learning accelerator and random access memory with a camera interface Download PDFInfo
- Publication number
- WO2021207237A1 WO2021207237A1 PCT/US2021/026008 US2021026008W WO2021207237A1 WO 2021207237 A1 WO2021207237 A1 WO 2021207237A1 US 2021026008 W US2021026008 W US 2021026008W WO 2021207237 A1 WO2021207237 A1 WO 2021207237A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- interface
- random access
- access memory
- matrix
- neural network
- Prior art date
Links
- 238000013135 deep learning Methods 0.000 title claims abstract description 146
- 238000013528 artificial neural network Methods 0.000 claims abstract description 218
- 239000011159 matrix material Substances 0.000 claims abstract description 94
- 238000012545 processing Methods 0.000 claims abstract description 91
- 238000000034 method Methods 0.000 claims abstract description 37
- 239000013598 vector Substances 0.000 claims description 123
- 239000000872 buffer Substances 0.000 claims description 35
- 230000003139 buffering effect Effects 0.000 claims description 3
- 229910052710 silicon Inorganic materials 0.000 claims description 3
- 239000010703 silicon Substances 0.000 claims description 3
- 230000008569 process Effects 0.000 abstract description 10
- 210000002569 neuron Anatomy 0.000 description 19
- 230000006870 function Effects 0.000 description 12
- 238000004891 communication Methods 0.000 description 10
- 238000003860 storage Methods 0.000 description 8
- 210000004027 cell Anatomy 0.000 description 6
- 230000004913 activation Effects 0.000 description 5
- 238000002059 diagnostic imaging Methods 0.000 description 4
- 238000010801 machine learning Methods 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 4
- 230000004044 response Effects 0.000 description 4
- 238000012549 training Methods 0.000 description 4
- 238000005265 energy consumption Methods 0.000 description 3
- 230000002093 peripheral effect Effects 0.000 description 3
- 230000000946 synaptic effect Effects 0.000 description 3
- 238000002604 ultrasonography Methods 0.000 description 3
- 241000699670 Mus sp. Species 0.000 description 2
- 238000009825 accumulation Methods 0.000 description 2
- 238000004364 calculation method Methods 0.000 description 2
- 238000004590 computer program Methods 0.000 description 2
- 238000003384 imaging method Methods 0.000 description 2
- 230000000644 propagated effect Effects 0.000 description 2
- 238000012546 transfer Methods 0.000 description 2
- 238000004458 analytical method Methods 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000000295 complement effect Effects 0.000 description 1
- 238000013527 convolutional neural network Methods 0.000 description 1
- 238000012937 correction Methods 0.000 description 1
- 125000004122 cyclic group Chemical group 0.000 description 1
- 238000013500 data storage Methods 0.000 description 1
- 201000010099 disease Diseases 0.000 description 1
- 208000037265 diseases, disorders, signs and symptoms Diseases 0.000 description 1
- 238000009826 distribution Methods 0.000 description 1
- 238000009510 drug design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000000977 initiatory effect Effects 0.000 description 1
- 238000012886 linear function Methods 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 229910044991 metal oxide Inorganic materials 0.000 description 1
- 150000004706 metal oxides Chemical class 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000001537 neural effect Effects 0.000 description 1
- 238000002360 preparation method Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 239000004065 semiconductor Substances 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
- 230000001960 triggered effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G11—INFORMATION STORAGE
- G11C—STATIC STORES
- G11C11/00—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor
- G11C11/21—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using electric elements
- G11C11/34—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using electric elements using semiconductor devices
-
- G—PHYSICS
- G11—INFORMATION STORAGE
- G11C—STATIC STORES
- G11C11/00—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor
- G11C11/54—Digital stores characterised by the use of particular electric or magnetic storage elements; Storage elements therefor using elements simulating biological cells, e.g. neuron
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/30—Arrangements for executing machine instructions, e.g. instruction decode
- G06F9/30003—Arrangements for executing specific machine instructions
- G06F9/30007—Arrangements for executing specific machine instructions to perform operations on data operands
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/30—Arrangements for executing machine instructions, e.g. instruction decode
- G06F9/30003—Arrangements for executing specific machine instructions
- G06F9/30007—Arrangements for executing specific machine instructions to perform operations on data operands
- G06F9/3001—Arithmetic instructions
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/30—Arrangements for executing machine instructions, e.g. instruction decode
- G06F9/30003—Arrangements for executing specific machine instructions
- G06F9/30007—Arrangements for executing specific machine instructions to perform operations on data operands
- G06F9/30036—Instructions to perform operations on packed data, e.g. vector, tile or matrix operations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/30—Arrangements for executing machine instructions, e.g. instruction decode
- G06F9/38—Concurrent instruction execution, e.g. pipeline or look ahead
- G06F9/3885—Concurrent instruction execution, e.g. pipeline or look ahead using a plurality of independent parallel functional units
- G06F9/3893—Concurrent instruction execution, e.g. pipeline or look ahead using a plurality of independent parallel functional units controlled in tandem, e.g. multiplier-accumulator
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F9/00—Arrangements for program control, e.g. control units
- G06F9/06—Arrangements for program control, e.g. control units using stored programs, i.e. using an internal store of processing equipment to receive or retain programs
- G06F9/46—Multiprogramming arrangements
- G06F9/50—Allocation of resources, e.g. of the central processing unit [CPU]
- G06F9/5005—Allocation of resources, e.g. of the central processing unit [CPU] to service a request
- G06F9/5027—Allocation of resources, e.g. of the central processing unit [CPU] to service a request the resource being a machine, e.g. CPUs, Servers, Terminals
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/06—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons
- G06N3/063—Physical realisation, i.e. hardware implementation of neural networks, neurons or parts of neurons using electronic means
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
- H04N23/60—Control of cameras or camera modules
- H04N23/617—Upgrading or updating of programs or applications for camera control
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
- H04N23/80—Camera processing pipelines; Components thereof
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
- H04N23/95—Computational photography systems, e.g. light-field imaging systems
- H04N23/951—Computational photography systems, e.g. light-field imaging systems by using two or more images to influence resolution, frame rate or aspect ratio
-
- H—ELECTRICITY
- H04—ELECTRIC COMMUNICATION TECHNIQUE
- H04N—PICTORIAL COMMUNICATION, e.g. TELEVISION
- H04N23/00—Cameras or camera modules comprising electronic image sensors; Control thereof
- H04N23/60—Control of cameras or camera modules
- H04N23/69—Control of means for changing angle of the field of view, e.g. optical zoom objectives or electronic zooming
Definitions
- At least some embodiments disclosed herein relate to integrated circuit devices in general and more particularly, but not limited to, integrated circuit devices having accelerators for Artificial Neural Networks (ANNs), such as ANNs configured through machine learning and/or deep learning.
- ANNs Artificial Neural Networks
- An Artificial Neural Network uses a network of neurons to process inputs to the network and to generate outputs from the network.
- each neuron in the network receives a set of inputs.
- Some of the inputs to a neuron may be the outputs of certain neurons in the network; and some of the inputs to a neuron may be the inputs provided to the neural network.
- the input/output relations among the neurons in the network represent the neuron connectivity in the network.
- each neuron can have a bias, an activation function, and a set of synaptic weights for its inputs respectively.
- the activation function may be in the form of a step function, a linear function, a log-sigmoid function, etc. Different neurons in the network may have different activation functions.
- each neuron can generate a weighted sum of its inputs and its bias and then produce an output that is the function of the weighted sum, computed using the activation function of the neuron.
- ANN model that includes the data representing the connectivity of the neurons in the network, as well as the bias, activation function, and synaptic weights of each neuron.
- a computing device can be configured to compute the output(s) of the network from a given set of inputs to the network.
- the inputs to an ANN network may be generated based on camera inputs; and the outputs from the ANN network may be the identification of an item, such as an event or an object.
- an ANN may be trained using a supervised method where the parameters in the ANN are adjusted to minimize or reduce the error between known outputs associated with or resulted from respective inputs and computed outputs generated via applying the inputs to the ANN.
- supervised learning/training methods include reinforcement learning and learning with error correction.
- an ANN may be trained using an unsupervised method where the exact outputs resulted from a given set of inputs is not known before the completion of the training.
- the ANN can be trained to classify an item into a plurality of categories, or data points into clusters.
- Deep learning uses multiple layers of machine learning to progressively extract features from input data. For example, lower layers can be configured to identify edges in an image; and higher layers can be configured to identify, based on the edges detected using the lower layers, items captured in the image, such as faces, objects, events, etc. Deep learning can be implemented via Artificial Neural Networks (ANNs), such as deep neural networks, deep belief networks, recurrent neural networks, and/or convolutional neural networks.
- ANNs Artificial Neural Networks
- Deep learning has been applied to many application fields, such as computer vision, speech/audio recognition, natural language processing, machine translation, bioinformatics, drug design, medical image processing, games, etc.
- FIG. 1 shows an integrated circuit device having a Deep Learning Accelerator and random access memory configured according to one embodiment.
- FIG. 2 shows a processing unit configured to perform matrix-matrix operations according to one embodiment.
- FIG. 3 shows a processing unit configured to perform matrix-vector operations according to one embodiment.
- FIG. 4 shows a processing unit configured to perform vector-vector operations according to one embodiment.
- FIG. 5 shows a Deep Learning Accelerator and random access memory configured to autonomously apply inputs to a trained Artificial Neural Network according to one embodiment.
- FIG. 6 shows an integrated circuit device having a Deep Learning Accelerator and random access memory configured with separate memory access connections according to one embodiment.
- FIG. 7 shows an integrated circuit device having a Deep Learning Accelerator and random access memory with a camera interface according to one embodiment.
- FIG. 8 shows a method implemented in an integrated circuit device according to one embodiment.
- At least some embodiments disclosed herein provide a general-purpose integrated circuit device configured to perform computations of Artificial Neural Networks (ANNs) with reduced energy consumption and computation time.
- the integrated circuit device includes a Deep Learning Accelerator (DLA) and random access memory.
- DLA Deep Learning Accelerator
- the integrated circuit device can be configured with separate connections for concurrent access to the random access memory.
- the integrated circuit device can have a camera interface configured to receive image data directly from one or more cameras as input to an Artificial Neural Network (ANN) implemented in the integrated circuit device.
- ANN Artificial Neural Network
- the Deep Learning Accelerator includes a set of general-purpose, programmable hardware computing logic that is specialized and/or optimized to perform parallel vector and/or matrix calculations, including but not limited to multiplication and accumulation of vectors and/or matrices.
- the Deep Learning Accelerator can include one or more Arithmetic-Logic Units (ALUs) to perform arithmetic and bitwise operations on integer binary numbers.
- ALUs Arithmetic-Logic Units
- the Deep Learning Accelerator is programmable via a set of instructions to perform the computations of an Artificial Neural Network (ANN).
- ANN Artificial Neural Network
- the granularity of the Deep Learning Accelerator (DLA) operating on vectors and matrices corresponds to the largest unit of vectors/matrices that can be operated upon during the execution of one instruction by the Deep Learning Accelerator (DLA).
- DLA Deep Learning Accelerator
- elements of vector/matrix operands can be operated upon by the Deep Learning Accelerator (DLA) in parallel to reduce execution time and/or energy consumption associated with memory/data access.
- the operations on vector/matrix operands of the granularity of the Deep Learning Accelerator (DLA) can be used as building blocks to implement computations on vectors/matrices of larger sizes.
- ANN Artificial Neural Network
- DLA Deep Learning Accelerator
- atomic computation capabilities of the Deep Learning Accelerator (DLA) in manipulating vectors and matrices of the granularity of the Deep Learning Accelerator (DLA) in response to instructions can be programmed to implement computations in an Artificial Neural Network (ANN).
- ANN Artificial Neural Network
- the Deep Learning Accelerator lacks some of the logic operation capabilities of a typical Central Processing Unit (CPU).
- the Deep Learning Accelerator (DLA) can be configured with sufficient logic units to process the input data provided to an Artificial Neural Network (ANN) and generate the output of the Artificial Neural Network (ANN) according to a set of instructions generated for the Deep Learning Accelerator (DLA).
- the Deep Learning Accelerator (DLA) can perform the computation of an Artificial Neural Network (ANN) with little or no help from a Central Processing Unit (CPU) or another processor.
- a conventional general purpose processor can also be configured as part of the Deep Learning Accelerator (DLA) to perform operations that cannot be implemented efficiently using the vector/matrix processing units of the Deep Learning Accelerator (DLA), and/or that cannot be performed by the vector/matrix processing units of the Deep Learning Accelerator (DLA).
- DLA Deep Learning Accelerator
- a typical Artificial Neural Network can be described/specified in a standard format (e.g., Open Neural Network Exchange (ONNX)).
- a compiler can be used to convert the description of the Artificial Neural Network (ANN) into a set of instructions for the Deep Learning Accelerator (DLA) to perform calculations of the Artificial Neural Network (ANN).
- the compiler can optimize the set of instructions to improve the performance of the Deep Learning Accelerator (DLA) in implementing the Artificial Neural Network (ANN).
- the Deep Learning Accelerator can have local memory, such as registers, buffers and/or caches, configured to store vector/matrix operands and the results of vector/matrix operations. Intermediate results in the registers can be pipelined/shifted in the Deep Learning Accelerator (DLA) as operands for subsequent vector/matrix operations to reduce time and energy consumption in accessing memory/data and thus speed up typical patterns of vector/matrix operations in implementing a typical Artificial Neural Network (ANN).
- the capacity of registers, buffers and/or caches in the Deep Learning Accelerator (DLA) is typically insufficient to hold the entire data set for implementing the computation of a typical Artificial Neural Network (ANN).
- a random access memory coupled to the Deep Learning Accelerator (DLA) is configured to provide an improved data storage capability for implementing a typical Artificial Neural Network (ANN).
- ANN Artificial Neural Network
- the Deep Learning Accelerator (DLA) loads data and instructions from the random access memory and stores results back into the random access memory.
- the communication bandwidth between the Deep Learning Accelerator (DLA) and the random access memory is configured to optimize or maximize the utilization of the computation power of the Deep Learning Accelerator (DLA).
- high communication bandwidth can be provided between the Deep Learning Accelerator (DLA) and the random access memory such that vector/matrix operands can be loaded from the random access memory into the Deep Learning Accelerator (DLA) and results stored back into the random access memory in a time period that is approximately equal to the time for the Deep Learning Accelerator (DLA) to perform the computations on the vector/matrix operands.
- the granularity of the Deep Learning Accelerator (DLA) can be configured to increase the ratio between the amount of computations performed by the Deep Learning Accelerator (DLA) and the size of the vector/matrix operands such that the data access traffic between the Deep Learning Accelerator (DLA) and the random access memory can be reduced, which can reduce the requirement on the communication bandwidth between the Deep Learning Accelerator (DLA) and the random access memory.
- DLA Deep Learning Accelerator
- the random access memory can include a portion configured to store input to the Artificial Neural Network (ANN) and another portion configured to store output from the Artificial Neural Network (ANN).
- One connection to the random access memory can be used by the Central Processing Unit (CPU) or another processor to access the output from the Artificial Neural Network (ANN), while concurrently another connection to the random access memory can be used by a Direct Memory Access (DMA) controller to store into the random access memory input data for the Artificial Neural Network (ANN).
- CPU Central Processing Unit
- DMA Direct Memory Access
- the Central Processing Unit can set up the Direct Memory Access (DMA) controller to write, into an input region of the random access memory, input data to be processed by the Artificial Neural Network (ANN).
- the completion of the Direct Memory Access (DMA) controller writing a set of input into the input region can trigger the Deep Learning Accelerator (DLA) to execute the instructions to implement Artificial Neural Network (ANN).
- DLA Deep Learning Accelerator
- ANN Artificial Neural Network
- the execution of the instructions results in the combination of the input with the matrices of the Artificial Neural Network (ANN) to generate output.
- the output is configured to be stored in another region of the random access memory for the Central Processing Unit (CPU).
- the model data of an Artificial Neural Network (ANN) can be stored in a further region of the random access memory.
- the model data can include the matrices identifying neural connectivity and synaptic weights of artificial neurons, states and/or properties of artificial neurons in the Artificial Neural Network (ANN).
- the model data can further include the instructions for the Deep Learning Accelerator (DLA) to implement the computation of the Artificial Neural Network (ANN).
- DLA Deep Learning Accelerator
- a compiler can convert a description of the Artificial Neural Network (ANN) into the model data stored in the random access memory.
- the Deep Learning Accelerator (DLA) and the Direct Memory Access (DMA) controller can process the input data using the Artificial Neural Network (ANN) without help from the Central Processing Unit (CPU).
- the output of the Artificial Neural Network (ANN) is automatically stored in the random access memory.
- the Central Processing Unit (CPU) can access the random access memory via a separate connection concurrently with the Direct Memory Access (DMA) controller providing input data to the Deep Learning Accelerator (DLA).
- a stream of input data to the Artificial Neural Network can be configured in the form of a sequence of input data sets.
- Each input data set is for a set of input to the Artificial Neural Network (ANN) during a time slot.
- DLA Deep Learning Accelerator
- DMA Direct Memory Access
- CPU Central Processing Unit
- ANN Artificial Neural Network
- the combination of the Deep Learning Accelerator (DLA), random access memory and the Direct Memory Access (DMA) controller can function as an independent supplier of results from an Artificial Neural Network (ANN) to the Central Processing Unit (CPU).
- the Central Processing Unit (CPU) can retrieve a set of output at a time when the output is needed.
- the Central Processing Unit (CPU) can instruct the Direct Memory Access (DMA) controller to pause its operations in supplying input to the Artificial Neural Network (ANN) when output from the Artificial Neural Network (ANN) is not required. Subsequently, when output from the Artificial Neural Network (ANN) is needed, the Central Processing Unit (CPU) can instruct the Direct Memory Access (DMA) controller to resume its operations of loading input data into the random access memory.
- ANN Artificial Neural Network
- DMA Direct Memory Access
- AN Ns Artificial Neural Networks
- digital cameras can be used to generate images for computer vision and/or autonomous driving, flying, navigation, etc.
- some sensor data can be converted into an image form for processing by an Artificial Neural Network (ANN).
- ANN Artificial Neural Network
- radar, lidar, ultrasound scanner, medical imaging equipment, etc. can generate images for analysis by Artificial Neural Networks (ANNs) to recognize and/or classify features, objects, diseases, etc.
- An integrated circuit device having a Deep Learning Accelerator and random access memory can be configured to include a camera interface to acquire image data as input to an Artificial Neural Network (ANN) implemented in the integrated circuit device.
- ANN Artificial Neural Network
- the camera interface can be implemented according to a Mobile Industry Processor Interface (MIPI) protocol to receive image data from an image sensor, a camera or another device that can generate images, such as radar, lidar, ultrasound scanner, medical imaging equipment.
- MIPI Mobile Industry Processor Interface
- the Mobile Industry Processor Interface (MIPI) protocol can include support of a camera command interface to control the operation of a camera, or an imaging device.
- the camera interface allows the integrated circuit device having a Deep Learning Accelerator and random access memory to control the receiving of image data as input for processing by an Artificial Neural Network (ANN).
- ANN Artificial Neural Network
- a camera or an image sensor/generator can stream its input data as images into the input region of the random access memory.
- the integrated circuit device automatically converts the input data according to an Artificial Neural Network (ANN) and stores the output from the Artificial Neural Network (ANN) in the random access memory. Since the integrated circuit device seamlessly acquires and converts the image data into Artificial Neural Network (ANN) outputs, data traffic to the Central Processing Unit can be dramatically reduced.
- ANN Artificial Neural Network
- the camera interface allows the integrated circuit device to be combined with an image generator as a smart sensor unit that automatically supplies the intelligent results from an Artificial Neural Network (ANN) in random access memory.
- ANN Artificial Neural Network
- the Central Processing Unit configures the integrated circuit device to operate, the computation tasks of acquiring input data and performing the computation involving the Artificial Neural Network (ANN) can be offloaded from the Central Processing Unit (CPU).
- the raw sensor/image data can be buffered in the random access memory for a period of time in a cyclic way such that if needed, the Central Processing Unit (CPU) can also access the raw sensor data within the period of time.
- FIG. 1 shows an integrated circuit device (101) having a Deep Learning Accelerator (103) and random access memory (105) configured according to one embodiment.
- the Deep Learning Accelerator (103) in FIG. 1 includes processing units (111), a control unit (113), and local memory (115).
- the controller unit (113) can use the processing units (111) to perform vector and matrix operations in accordance with instructions. Further, the controller unit (113) can load instructions and operands from the random access memory (105) through a memory interface (117) and a high speed/bandwidth connection (119).
- the integrated circuit device (101) is configured to be enclosed within an integrated circuit package with pins or contacts for a memory controller interface (107).
- the memory controller interface (107) is configured to support a standard memory access protocol such that the integrated circuit device (101) appears to a typical memory controller in a way same as a conventional random access memory device having no Deep Learning Accelerator (DLA) (103).
- DLA Deep Learning Accelerator
- a memory controller external to the integrated circuit device (101) can access, using a standard memory access protocol through the memory controller interface (107), the random access memory (105) in the integrated circuit device (101).
- the integrated circuit device (101) is configured with a high bandwidth connection (119) between the random access memory (105) and the Deep Learning Accelerator (DLA) (103) that are enclosed within the integrated circuit device (101).
- the bandwidth of the connection (119) is higher than the bandwidth of the connection (109) between the random access memory (105) and the memory controller interface (107).
- both the memory controller interface (107) and the memory interface (117) are configured to access the random access memory (105) via a same set of buses or wires.
- the bandwidth to access the random access memory (105) is shared between the memory interface (117) and the memory controller interface (107).
- the memory controller interface (107) and the memory interface (117) are configured to access the random access memory (105) via separate sets of buses or wires.
- the random access memory (105) can include multiple sections that can be accessed concurrently via the connection (119). For example, when the memory interface (117) is accessing a section of the random access memory (105), the memory control interface (107) can concurrently access another section of the random access memory (105).
- the different sections can be configured on different integrated circuit dies and/or different planes/banks of memory cells; and the different sections can be accessed in parallel to increase throughput in accessing the random access memory (105).
- the memory controller interface (107) is configured to access one data unit of a predetermined size at a time; and the memory interface (117) is configured to access multiple data units, each of the same predetermined size, at a time.
- the random access memory (105) and the integrated circuit device (101) are configured on different integrated circuit dies configured within a same integrated circuit package. Further, the random access memory (105) can be configured on one or more integrated circuit dies that allows parallel access of multiple data elements concurrently.
- the number of data elements of a vector or matrix that can be accessed in parallel over the connection (119) corresponds to the granularity of the Deep Learning Accelerator (DLA) operating on vectors or matrices.
- DLA Deep Learning Accelerator
- the connection (119) is configured to load or store the same number, or multiples of the number, of elements via the connection (119) in parallel.
- the data access speed of the connection (119) can be configured based on the processing speed of the Deep Learning Accelerator (DLA) (103).
- DLA Deep Learning Accelerator
- the control unit (113) can execute an instruction to operate on the data using the processing units (111) to generate output.
- the access bandwidth of the connection (119) allows the same amount of data and instructions to be loaded into the local memory (115) for the next operation and the same amount of output to be stored back to the random access memory (105).
- the memory interface (117) can offload the output of a prior operation into the random access memory (105) from, and load operand data and instructions into, another portion of the local memory (115).
- DLA Deep Learning Accelerator
- the random access memory (105) can be used to store the model data of an Artificial Neural Network (ANN) and to buffer input data for the Artificial Neural Network (ANN).
- the model data does not change frequently.
- the model data can include the output generated by a compiler for the Deep Learning Accelerator (DLA) to implement the Artificial Neural Network (ANN).
- the model data typically includes matrices used in the description of the Artificial Neural Network (ANN) and instructions generated for the Deep Learning Accelerator (DLA) (103) to perform vector/matrix operations of the Artificial Neural Network (ANN) based on vector/matrix operations of the granularity of the Deep Learning Accelerator (DLA) (103).
- the instructions operate not only on the vector/matrix operations of the Artificial Neural Network (ANN), but also on the input data for the Artificial Neural Network (ANN).
- the control unit (113) of the Deep Learning Accelerator (DLA) (103) can automatically execute the instructions for the Artificial Neural Network (ANN) to generate an output of the Artificial Neural Network (ANN).
- the output is stored into a predefined region in the random access memory (105).
- the Deep Learning Accelerator (DLA) (103) can executed the instructions without help from a Central Processing Unit (CPU).
- CPU Central Processing Unit
- the logic circuit of the Deep Learning Accelerator (DLA) (103) can be implemented via Complementary Metal Oxide Semiconductor (CMOS).
- CMOS Complementary Metal Oxide Semiconductor
- the technique of CMOS Under the Array (CUA) of memory cells of the random access memory (105) can be used to implement the logic circuit of the Deep Learning Accelerator (DLA) (103), including the processing units (111) and the control unit (113).
- the technique of CMOS in the Array of memory cells of the random access memory (105) can be used to implement the logic circuit of the Deep Learning Accelerator (DLA) (103).
- the Deep Learning Accelerator (DLA) (103) and the random access memory (105) can be implemented on separate integrated circuit dies and connected using Through-Silicon Vias (TSV) for increased data bandwidth between the Deep Learning Accelerator (DLA) (103) and the random access memory (105).
- TSV Through-Silicon Vias
- the Deep Learning Accelerator (DLA) (103) can be formed on an integrated circuit die of a Field-Programmable Gate Array (FPGA) or Application Specific Integrated circuit (ASIC).
- FPGA Field-Programmable Gate Array
- ASIC Application Specific Integrated circuit
- the Deep Learning Accelerator (DLA) (103) and the random access memory (105) can be configured in separate integrated circuit packages and connected via multiple point-to-point connections on a printed circuit board (PCB) for parallel communications and thus increased data transfer bandwidth.
- DLA Deep Learning Accelerator
- PCB printed circuit board
- the random access memory (105) can be volatile memory or non-volatile memory, or a combination of volatile memory and non-volatile memory.
- non-volatile memory include flash memory, memory cells formed based on negative- and (NAND) logic gates, negative-or (NOR) logic gates, Phase-Change Memory (PCM), magnetic memory (MRAM), resistive random-access memory, cross point storage and memory devices.
- a cross point memory device can use transistor-less memory elements, each of which has a memory cell and a selector that are stacked together as a column.
- Memory element columns are connected via two lays of wires running in perpendicular directions, where wires of one lay run in one direction in the layer that is located above the memory element columns, and wires of the other lay run in another direction and are located below the memory element columns.
- Each memory element can be individually selected at a cross point of one wire on each of the two layers.
- Cross point memory devices are fast and non-volatile and can be used as a unified memory pool for processing and storage.
- Further examples of non-volatile memory include Read-Only Memory (ROM), Programmable Read-Only Memory (PROM), Erasable Programmable Read-Only Memory (EPROM) and Electronically Erasable Programmable Read-Only Memory (EEPROM) memory, etc.
- Examples of volatile memory include Dynamic Random-Access Memory (DRAM) and Static Random-Access Memory (SRAM).
- non-volatile memory can be configured to implement at least a portion of the random access memory (105).
- the non-volatile memory in the random access memory (105) can be used to store the model data of an Artificial Neural Network (ANN).
- ANN Artificial Neural Network
- the non-volatile memory can be programmable/rewritable.
- the model data of the Artificial Neural Network (ANN) in the integrated circuit device (101) can be updated or replaced to implement an update Artificial Neural Network (ANN), or another Artificial Neural Network (ANN).
- the processing units (111) of the Deep Learning Accelerator (DLA) (103) can include vector-vector units, matrix-vector units, and/or matrix-matrix units. Examples of units configured to perform for vector-vector operations, matrix-vector operations, and matrix-matrix operations are discussed below in connection with
- FIG. 2 shows a processing unit (121) configured to perform matrix-matrix operations according to one embodiment.
- the matrix-matrix unit (121 ) of FIG. 2 can be used as one of the processing units (111 ) of the Deep Learning Accelerator (DLA) (103) of FIG. 1.
- DLA Deep Learning Accelerator
- the matrix-matrix unit (121) includes multiple kernel buffers (131 to 133) and multiple the maps banks (151 to 153). Each of the maps banks (151 to 153) stores one vector of a matrix operand that has multiple vectors stored in the maps banks (151 to 153) respectively; and each of the kernel buffers (131 to 133) stores one vector of another matrix operand that has multiple vectors stored in the kernel buffers (131 to 133) respectively.
- the matrix-matrix unit (121) is configured to perform multiplication and accumulation operations on the elements of the two matrix operands, using multiple matrix-vector units (141 to 143) that operate in parallel. [0063]
- a cross bar (123) connects the maps banks (151 to 153) to the matrix- vector units (141 to 143).
- the same matrix operand stored in the maps bank (151 to 153) is provided via the crossbar (123) to each of the matrix-vector units (141 to 143); and the matrix-vector units (141 to 143) receives data elements from the maps banks (151 to 153) in parallel.
- Each of the kernel buffers (131 to 133) is connected to a respective one in the matrix-vector units (141 to 143) and provides a vector operand to the respective matrix-vector unit.
- the matrix-vector units (141 to 143) operate concurrently to compute the operation of the same matrix operand, stored in the maps banks (151 to 153) multiplied by the corresponding vectors stored in the kernel buffers (131 to 133).
- the matrix-vector unit (141) performs the multiplication operation on the matrix operand stored in the maps banks (151 to 153) and the vector operand stored in the kernel buffer (131), while the matrix-vector unit (143) is concurrently performing the multiplication operation on the matrix operand stored in the maps banks (151 to 153) and the vector operand stored in the kernel buffer (133).
- Each of the matrix-vector units (141 to 143) in FIG. 2 can be implemented in a way as illustrated in FIG. 3.
- FIG. 3 shows a processing unit (141) configured to perform matrix-vector operations according to one embodiment.
- the matrix-vector unit (141 ) of FIG. 3 can be used as any of the matrix-vector units in the matrix-matrix unit (121) of FIG. 2.
- each of the maps banks (151 to 153) stores one vector of a matrix operand that has multiple vectors stored in the maps banks (151 to 153) respectively, in a way similar to the maps banks (151 to 153) of FIG. 2.
- the crossbar (123) in FIG. 3 provides the vectors from the maps banks (151) to the vector-vector units (161 to 163) respectively.
- a same vector stored in the kernel buffer (131) is provided to the vector-vector units (161 to 163).
- the vector-vector units (161 to 163) operate concurrently to compute the operation of the corresponding vector operands, stored in the maps banks (151 to 153) respectively, multiplied by the same vector operand that is stored in the kernel buffer (131 ).
- the vector-vector unit (161) performs the multiplication operation on the vector operand stored in the maps bank (151) and the vector operand stored in the kernel buffer (131), while the vector-vector unit (163) is concurrently performing the multiplication operation on the vector operand stored in the maps bank (153) and the vector operand stored in the kernel buffer (131).
- the matrix-vector unit (141) of FIG. 3 can use the maps banks (151 to 153), the crossbar (123) and the kernel buffer (131) of the matrix-matrix unit (121).
- Each of the vector-vector units (161 to 163) in FIG. 3 can be implemented in a way as illustrated in FIG. 4.
- FIG. 4 shows a processing unit (161 ) configured to perform vector-vector operations according to one embodiment.
- the vector-vector unit (161 ) of FIG. 4 can be used as any of the vector-vector units in the matrix-vector unit (141 ) of FIG. 3.
- the vector-vector unit (161) has multiple multiply-accumulate (MAC) units (171 to 173).
- Each of the multiply-accumulate (MAC) units (171 to 173) can receive two numbers as operands, perform multiplication of the two numbers, and add the result of the multiplication to a sum maintained in the multiply- accumulate (MAC) unit.
- Each of the vector buffers (181 and 183) stores a list of numbers.
- a pair of numbers, each from one of the vector buffers (181 and 183), can be provided to each of the multiply-accumulate (MAC) units (171 to 173) as input.
- the multiply- accumulate (MAC) units (171 to 173) can receive multiple pairs of numbers from the vector buffers (181 and 183) in parallel and perform the multiply-accumulate (MAC) operations in parallel.
- the outputs from the multiply-accumulate (MAC) units (171 to 173) are stored into the shift register (175); and an accumulator (177) computes the sum of the results in the shift register (175).
- the vector-vector unit (161) of FIG. 4 is implemented in a matrix- vector unit (141) of FIG. 3, the vector-vector unit (161 ) can use a maps bank (e.g.,
- the vector buffers (181 and 183) can have a same length to store the same number/count of data elements.
- the length can be equal to, or the multiple of, the count of multiply-accumulate (MAC) units (171 to 173) in the vector-vector unit (161).
- the communication bandwidth of the connection (119) between the Deep Learning Accelerator (DLA) (103) and the random access memory (105) is sufficient for the matrix-matrix unit (121 ) to use portions of the random access memory (105) as the maps banks (151 to 153) and the kernel buffers (131 to 133).
- the maps banks (151 to 153) and the kernel buffers (131 to 133) are implemented in a portion of the local memory (115) of the Deep Learning Accelerator (DLA) (103).
- the communication bandwidth of the connection (119) between the Deep Learning Accelerator (DLA) (103) and the random access memory (105) is sufficient to load, into another portion of the local memory (115), matrix operands of the next operation cycle of the matrix-matrix unit (121), while the matrix-matrix unit (121) is performing the computation in the current operation cycle using the maps banks (151 to 153) and the kernel buffers (131 to 133) implemented in a different portion of the local memory (115) of the Deep Learning Accelerator (DLA) (103).
- FIG. 5 shows a Deep Learning Accelerator and random access memory configured to autonomously apply inputs to a trained Artificial Neural Network according to one embodiment.
- An Artificial Neural Network (ANN) (201 ) that has been trained through machine learning (e.g., deep learning) can be described in a standard format (e.g., Open Neural Network Exchange (ONNX)).
- the description of the trained ANN (201) in the standard format identifies the properties of the artificial neurons and their connectivity.
- a Deep Learning Accelerator (DLA) compiler (203) converts trained ANN (201) by generating instructions (205) for a Deep Learning Accelerator (DLA) (103) and matrices (207) corresponding to the properties of the artificial neurons and their connectivity.
- the instructions (205) and the matrices (207) generated by the DLA compiler (203) from the trained ANN (201 ) can be stored in random access memory (105) for the Deep Learning Accelerator (DLA) (103).
- the random access memory (105) and the Deep Learning Accelerator (DLA) (103) can be connected via a high bandwidth connection (119) in a way as in the integrated circuit device (101 ) of FIG. 1.
- the autonomous computation of FIG. 5 based on the instructions (205) and the matrices (207) can be implemented in the integrated circuit device (101) of FIG. 1.
- the random access memory (105) and the Deep Learning Accelerator (DLA) (103) can be configured on a printed circuit board with multiple point to point serial buses running in parallel to implement the connection (119).
- the application of the trained ANN (201) to process an input (211 ) to the trained ANN (201 ) to generate the corresponding output (213) of the trained ANN (213) can be triggered by the presence of the input (211 ) in the random access memory (105), or another indication provided in the random access memory (105).
- the Deep Learning Accelerator (DLA) (103) executes the instructions (205) to combine the input (211 ) and the matrices (207).
- the execution of the instructions (205) can include the generation of maps matrices for the maps banks (151 to 153) of one or more matrix-matrix units (e.g., 121) of the Deep Learning Accelerator (DLA) (103).
- the inputs to ANN (211 ) is in the form of an initial maps matrix. Portions of the initial maps matrix can be retrieved from the random access memory (105) as the matrix operand stored in the maps banks (151 to 153) of a matrix-matrix unit (121).
- the DLA instructions (205) also include instructions for the Deep Learning Accelerator (DLA) (103) to generate the initial maps matrix from the input (211 ).
- DLA Deep Learning Accelerator
- the Deep Learning Accelerator (DLA) (103) loads matrix operands into the kernel buffers (131 to 133) and maps banks (151 to 153) of its matrix-matrix unit (121).
- the matrix-matrix unit (121) performs the matrix computation on the matrix operands.
- the DLA instructions (205) break down matrix computations of the trained ANN (201 ) according to the computation granularity of the Deep Learning Accelerator (DLA) (103) (e.g., the sizes/dimensions of matrices that loaded as matrix operands in the matrix-matrix unit (121)) and applies the input feature maps to the kernel of a layer of artificial neurons to generate output as the input for the next layer of artificial neurons.
- the Deep Learning Accelerator (DLA) (103) Upon completion of the computation of the trained ANN (201 ) performed according to the instructions (205), the Deep Learning Accelerator (DLA) (103) stores the output (213) of the ANN (201 ) at a pre-defined location in the random access memory (105), or at a location specified in an indication provided in the random access memory (105) to trigger the computation.
- DLA Deep Learning Accelerator
- an external device connected to the memory controller interface (107) can write the input (211) into the random access memory (105) and trigger the autonomous computation of applying the input (211) to the trained ANN (201) by the Deep Learning Accelerator (DLA) (103). After a period of time, the output (213) is available in the random access memory (105); and the external device can read the output (213) via the memory controller interface (107) of the integrated circuit device (101).
- DLA Deep Learning Accelerator
- a predefined location in the random access memory (105) can be configured to store an indication to trigger the autonomous execution of the instructions (205) by the Deep Learning Accelerator (DLA) (103).
- the indication can optionally include a location of the input (211 ) within the random access memory (105).
- the external device can retrieve the output generated during a previous run of the instructions (205), and/or store another set of input for the next run of the instructions (205).
- a further predefined location in the random access memory (105) can be configured to store an indication of the progress status of the current run of the instructions (205). Further, the indication can include a prediction of the completion time of the current run of the instructions (205) (e.g., estimated based on a prior run of the instructions (205)). Thus, the external device can check the completion status at a suitable time window to retrieve the output (213).
- the random access memory (105) is configured with sufficient capacity to store multiple sets of inputs (e.g., 211 ) and outputs (e.g., 213). Each set can be configured in a predetermined slot/area in the random access memory (105).
- the Deep Learning Accelerator (DLA) (103) can execute the instructions (205) autonomously to generate the output (213) from the input (211) according to matrices (207) stored in the random access memory (105) without helps from a processor or device that is located outside of the integrated circuit device (101).
- random access memory (103) of a computing device e.g., 101
- the computing device (e.g., 101) can have processing units (e.g., 111) configured to perform at least computations on matrix operands, such as a matrix operand stored in maps banks (151 to 153) and a matrix operand stored in kernel buffers (131 to 133).
- processing units e.g., 111
- matrix operands such as a matrix operand stored in maps banks (151 to 153) and a matrix operand stored in kernel buffers (131 to 133).
- the computing device e.g., 101
- the computing device can be enclosed within an integrated circuit package; and a set of connections can connect the interface (107) to the memory controller that is located outside of the integrated circuit package.
- Instructions (205) executable by the processing units e.g., 111
- Matrices (207) of an Artificial Neural Network (201 ) can be written into the random access memory (105) through the interface (107).
- the matrices (207) identify the property and/or state of the Artificial Neural Network (201 ).
- At least a portion of the random access memory (105) is non volatile and configured to store the instructions (205) and the matrices (07) of the Artificial Neural Network (201 ).
- First input (211) to the Artificial Neural Network can be written into the random access memory (105) through the interface (107).
- An indication is provided in the random access memory (105) to cause the processing units (111) to start execution of the instructions (205).
- the processing units (111) execute the instructions to combine the first input (211) with the matrices (207) of the Artificial Neural Network (201 ) to generate first output (213) from the Artificial Neural Network (201 ) and store the first output (213) in the random access memory (105).
- the indication can be an address of the first input (211 ) in the random access memory (105); and the indication can be stored a predetermined location in the random access memory (105) to cause the initiation of the execution of the instructions (205) for the input (211 ) identified by the address.
- the indication can also include an address for storing the output (213).
- the first output (213) can be read, through the interface (107), from the random access memory (105).
- the computing device e.g., 101
- the computing device can have a Deep Learning Accelerator (103) formed on a first integrated circuit die and the random access memory (105) formed on one or more second integrated circuit dies.
- the connection (119) between the first integrated circuit die and the one or more second integrated circuit dies can include Through-Silicon Vias (TSVs) to provide high bandwidth for memory access.
- TSVs Through-Silicon Vias
- a description of the Artificial Neural Network (201 ) can be converted using a compiler (203) into the instructions (205) and the matrices (207).
- the combination of the instructions (205) and the matrices (207) stored in the random access memory (105) and the Deep Learning Accelerator (103) provides an autonomous implementation of the Artificial Neural Network (201) that can automatically convert input (211 ) to the Artificial Neural Network (201 ) to its output (213).
- the Deep Learning Accelerator (103) executes the instructions (205) to generate the first output (213) from the first input (211) according to the matrices (207) of the Artificial Neural Network (201)
- the second input to Artificial Neural Network (201) can be written into the random access memory (105) through the interface (107) at an alternative location.
- an indication can be provided in the random access memory to cause the Deep Learning Accelerator (103) to again start the execution of the instructions and generate second output from the second input.
- the first output (213) can be read from the random access memory (105) through the interface (107); and a further input can be written into the random access memory to replace the first input (211), or written at a different location. The process can be repeated for a sequence of inputs.
- the Deep Learning Accelerator (103) can include at least one matrix- matrix unit (121) that can execute an instruction on two matrix operands.
- the two matrix operands can be a first matrix and a second matrix.
- Each of two matrices has a plurality of vectors.
- the matrix-matrix unit (121 ) can include a plurality of matrix- vector units (141 to 143) configured to operate in parallel.
- Each of the matrix-vector units (141 to 143) are configured to operate, in parallel with other matrix-vector units, on the first matrix and one vector from second matrix.
- each of the matrix- vector units (141 to 143) can have a plurality of vector-vector units (161 to 163) configured to operate in parallel.
- Each of the vector-vector units (161 to 163) is configured to operate, in parallel with other vector-vector units, on a vector from the first matrix and a common vector operand of the corresponding matrix-vector unit. Further, each of the vector-vector units (161 to 163) can have a plurality of multiply- accumulate units (171 to 173) configured to operate in parallel.
- the Deep Learning Accelerator (103) can have local memory (115) and a control unit (113) in addition to the processing units (111).
- the control unit (113) can load instructions (205) and matrix operands (e.g., 207) from the random access memory (105) for execution by the processing units (111).
- the local memory can cache matrix operands used by the matrix-matrix unit.
- the connection (119) can be configured with a bandwidth sufficient to load a set of matrix operands from the random access memory (105) to the local memory (115) during a time period in which the matrix-matrix unit performs operations on two other matrix operands. Further, during the time period, the bandwidth is sufficient to store a result, generated by the matrix-matrix unit (121) in a prior instruction execution, from the local memory (115) to the random access memory (105).
- FIG. 6 shows an integrated circuit device (101) having a Deep Learning Accelerator (103) and random access memory (105) configured with separate memory access connections according to one embodiment.
- the Deep Learning Accelerator (103), the random access memory (105), and the connection (119) of the integrated circuit device (101) of FIG. 6 can be configured in a way similar to those illustrated in FIG. 1 and/or FIG. 5.
- the integrated circuit device (101) of FIG. 6 has two external interfaces (106 and 107) that can be used to access the random access memory (105) concurrently.
- the random access memory (105) can have at least two portions that can be accessed concurrently and independently from reach other. Such portions can be configured on separate integrated circuit dies, or in separate planes or blocks of memory cells form on a same integrated circuit die. One portion is configured to store input (211) to the Artificial Neural Network (ANN) (201); and another portion is configured to store output (213) from the Artificial Neural Network (ANN) (201).
- the two external interfaces (106 and 107) configured with separate connections (108 and 109) to the portions for the input (211) and output (213) respectively.
- different devices external to the integrated circuit (101) can use the separate external interfaces (106 and 107) of the integrated circuit device (101 ) to access its random access memory (105) concurrently.
- a memory controller of a Central Processing Unit can be connected to the CPU memory interface (107) to read prior output from the Artificial Neural Network (ANN) (201), while a Direct Memory Access (DMA) controller can be connected to the DMA controller interface (106) to write next input to the Artificial Neural Network (ANN) (201) concurrently.
- the connections (108 and 109) have separate sets of buses or wires.
- the external interfaces (106 and 107) do not share buses or wires in accessing the different portions of the random access memory (105) for the input (211) and the output (213).
- an access controller is configured to use separate buffers for the interfaces (106 and 107) and use the high bandwidth connection (119) to transfer data between the random access memory (1050 and the buffers for the interfaces (106 and 107) such that the interfaces (106 and 107) can service write and read requests concurrently. Since the bandwidth of the connection (119) is substantially higher than the bandwidth used by the connections (108 and 109) to the external interfaces (106 and 107) of the integrated circuit device (101 ), a small portion of the bandwidth can be allocated to the connections (108 and 109).
- the interfaces (106 and 107) can be connected to the memory interface (117) of the Deep Learning Accelerator (DLA) (103) to access the random access memory (105) via the connection (119).
- DLA Deep Learning Accelerator
- the memory interface (117) of the Deep Learning Accelerator (DLA) (103), the interface (107) to the memory controller of the Central Processing Unit (CPU), and the interface (106) can be configured to access the random access memory (105) concurrently.
- the random access memory (105) can be configured with at multiple sets of input/output memory. Each set can be selectively configured to service the memory interface (117) or service the external interfaces (106 and 107).
- the connection (119) allows the Deep Learning Accelerator (DLA) (103) to access input (e.g., 211) stored in the set and to store output (e.g., 213) from the Artificial Neural Network (ANN) (201) to the set.
- DLA Deep Learning Accelerator
- ANN Artificial Neural Network
- the input (e.g., 211 ) and the output (e.g., 213) in different memory regions can be accessed concurrently by the separate external interfaces (106 and 107). While one set of input/output memory is allocated for the Deep Learning Accelerator (DLA) (103) to process a set of input (e.g., 211) and generate a corresponding set of output (e.g., 213), one or more sets of input/output memory can be concurrently made accessible to the external interfaces (106 and 107).
- DLA Deep Learning Accelerator
- an integrated circuit device (101) stores matrices (207) of an Artificial Neural Network (201) and instructions (205).
- the instructions (205) are executable by at least one processing unit (111) enclosed within the integrated circuit device (101) to implement the Artificial Neural Network (201) using the matrices (207).
- the integrated circuit device (101), or the alternatively packaged computing device on a printed circuit board, has random access memory.
- the random access memory (105) enclosed within the integrated circuit device (101 ) stores first input to the Artificial Neural Network (201 ); and the at least one processing unit (111 ) is caused or used to execute the instructions (205) in generating first output from the first input stored in the random access memory (105). The first output is stored in the random access memory (105).
- the random access memory (105) enclosed within the integrated circuit device (101) further stores second input to the Artificial Neural Network (201); and the at least one processing unit (111 ) is caused or used to execute the instructions (205) in generating second output from the second input stored in the random access memory (105).
- an external device e.g., a direct memory access controller
- another external device e.g., a Central Processing Unit
- the writing of the third input can be performed through the first interface (106) connected to a direct memory access controller; and the reading of the first output can be performed through the second interface (107) connected to a Central Processing Unit.
- the integrated circuit device (101) can be enclosed within an integrated circuit package and has a Deep Learning Accelerator (103) with processing units (111), a control unit (113) and local memory (115).
- the processing units (111) include at least a matrix-matrix unit (121) configured to execute an instruction having two matrix operands.
- the matrix-matrix unit (121) includes a plurality of matrix-vector units (141 to 143) configured to operate in parallel.
- Each of the matrix-vector units (141 to 143) includes a plurality of vector-vector units (161 to 163) configured to operate in parallel.
- Each of the vector-vector units (161 to 163) includes a plurality of multiply-accumulate units (171 to 173) configured to operate in parallel.
- a compiler (203) can be used to convert a description of the Artificial Neural Network (201 ) into the instructions (205) and the matrices (207) to implement the Artificial Neural Network (101) using the Deep Learning Accelerator (103).
- the reading of the first output through the second interface (107), the writing of the third input through the first interface (106), the Deep Learning Accelerator (103) reading a portion of the second input, and the Deep Learning Accelerator (103) writing a portion of the second output into the random access memory can be performed in parallel and concurrently.
- the random access memory (105) can have multiple portions that are capable of being used concurrently and independent from each other.
- a first portion is configured to store the first output from the Artificial Neural Network (201 ); a second portion configured to store third input to the Artificial Neural Network (201); a third portion configured to store the second output from the Artificial Neural Network (201); and a fourth portion configured to store the second input to the Artificial Neural Network (201 ).
- the third and fourth portions are being used by the Deep Learning Accelerator 103) in execution of the instructions (205), the first interface and the second interface can be connected concurrently to the first portion and second portion respectively.
- the different portions can be configured on separate integrated circuit dies (or planes or blocks) that can operate independent from each other in parallel.
- the first interface and the second interface can share no connections to the first portion and the second portion.
- the integrated circuit device (101) can be enclosed within a single integrated circuit package, with a first set of connectors configured to couple the first interface (106) to the direct memory access controller and a second set of connectors configured to couple the second interface (107) to the Central Processing Unit.
- the Deep Learning Accelerator (103) has a memory interface (117) with a high bandwidth connection (119) to the random access memory (105); and the first interface (106) and the second interface (107) are connected to the random access memory (105) via the memory interface (117) of the Deep Learning Accelerator (117).
- FIG. 7 shows an integrated circuit device (101) having a Deep Learning Accelerator (103) and random access memory (105) with a camera interface (221) according to one embodiment.
- the Deep Learning Accelerator (103), the random access memory (105), and the connection (119) of the integrated circuit device (101) of FIG. 7 can be configured in a way similar to those illustrated in FIG. 1, FIG. 5, and/or FIG. 6
- the integrated circuit device (101) of FIG. 7 has at least two external interfaces (221 and 107) that can be used concurrently, including a CPU memory interface (107) and a camera interface (221 ).
- the CPU memory interface (107) in the integrated circuit device (101) of FIG. 7 is configured to provide for a connection (229) to a memory controller (227) of a Central Processing Unit (CPU) (225), similar to the CPU memory interface (107) in the integrated circuit device (101 ) of FIG. 6
- CPU Central Processing Unit
- the CPU memory interface (107) is implemented in the Deep Learning Accelerator (103).
- the logic circuit of the CPU memory interface (107) can be formed on the integrated circuit die of the Deep Learning Accelerator (103); and the CPU memory interface (107) accesses the random access memory (105) using the memory interface (117) of the Deep Learning Accelerator (103).
- the CPU memory interface (107) can be separate from the Deep Learning Accelerator (103) and be implemented in a way as illustrated in FIG. 6.
- the CPU memory interface (107) can be configured to access the random access memory (105) using a connection (109) that is separate from the connection (119) between the Deep Learning Accelerator (103) and the random access memory (105).
- the integrated circuit device (101) of FIG. 6 can also be modified to have the logic circuit of the CPU memory interface (107) and/or the DMA controller interface (106) implemented in the Deep Learning Accelerator (DLA) (103), in a way similar to the CPU memory interface (107) being implemented in the Deep Learning Accelerator (103) of FIG. 7.
- the camera interface (221) in the integrated circuit device (101 ) of FIG. 7 is configured to provide a connection (228) to one or more devices containing or employing image sensors or image generators, such as a camera (223), radar, lidar, medical imaging equipment, etc.
- connection (228) to the camera (223) can be implemented in accordance with a standard for a Mobile Industry Processor Interface (MIPI) protocol, including a MIPI Camera Serial Interface (CSI) protocol.
- MIPI Mobile Industry Processor Interface
- CSI MIPI Camera Serial Interface
- the connection (228) can be used to control the imaging operations of the camera (223) (or another image generator) and to obtain image data from the camera (223) (or another image generator) as input (211 ) to the Artificial Neural Network (201 ).
- the connection (228) is a serial bus in accordance with a MIPI CSI- 2 standard that supports high-performance applications, including 1080p, 4K, 8K and beyond video, and high-resolution photography.
- the camera interface (221) can be used to control the operations of the camera (223) in generating image data.
- the camera interface (221) can be used to send commands to the camera (223) to adjust the resolution of image data provided by the camera (223).
- the commands sent via the camera interface (221 ) to the camera (223) can adjust the frame rate and/or exposure time of image data provided by the camera (223).
- the Deep Learning Accelerator (DLA) (103) can use the camera interface (221 ) to start or stop the operations of the camera (223) in capturing and/or transferring images.
- DLA Deep Learning Accelerator
- the camera interface (221) can be used to optionally issue commands to digitally zoom in an area of an interest and thus reduce the amount of data to be transferred over the connection (228) and the amount of computation to be performed by the integrated circuit device (101 ) on the image data received from the camera (223).
- the camera interface (221) can be used to optionally issue commands to adjust the view point and/or field of view of the camera (223).
- FIG. 7 illustrates an implementation in which the camera interface (221) is configured in the Deep Learning Accelerator (103).
- the camera interface (221 ) is formed on the integrated circuit die of the Deep Learning Accelerator (103); and the camera interface (107) can access the random access memory (105) using the memory interface (117) of the Deep Learning Accelerator (103).
- the camera interface (221) can receive image from the camera (223) through the connection (228) and buffers the image data in the random access memory (105) as input (211 ) to the Artificial Neural Network (201 ).
- the Deep Learning Accelerator (103) can execute the instructions (205) to generate the output (213), in a way similar to the device of FIG. 1, FIG. 5, and/or FIG. 6.
- the instructions (205) can optionally include instructions for the control unit (113) to operate the camera interface (221) and obtain image data, from the camera (223) via the connection (228) through the camera interface (221), as the input (211) to the Artificial Neural Network (201).
- the camera interface (221) is configured in the Deep Learning Accelerator (103).
- the camera interface (221) can be separate from the Deep Learning Accelerator (103) and be configured to access the random access memory (105) using a connection that is separate from the connection (119) between the Deep Learning Accelerator (103) and the random access memory (105), in a way similar to the DMA controller interface (106) being separate from the Deep Learning Accelerator (103) in FIG. 6.
- the integrated circuit devices (101) of FIG. 6 and FIG. 7 can be modified to have three external interfaces: a CPU memory interface (107), a DMA controller interface (106), and a camera interface (221).
- the camera interface (221 ) can include a direct memory access controller to load image data into the random access memory (105) from the camera (223), e.g., in parallel with the Central Processing Unit (225) retrieving output (e.g., 213) from the random access memory (105).
- the camera interface (221 ) and the DMA controller interface (106) can optionally share a portion of logic circuit for supporting the operations of direct memory access controllers.
- FIG. 8 shows a method implemented in an integrated circuit device according to one embodiment.
- the method of FIG. 8 can be implemented in the integrated circuit device (101) of FIG. 7.
- the method of FIG. 8 can also be implemented in an alternatively packaged computing device similar to that illustrated in FIG. 5 but configured on a printed circuit board.
- an integrated circuit device (101 ) stores matrices (207) of an Artificial Neural Network (201 ) and instructions (205).
- the instructions (205) are executable by at least one processing unit (111 ) enclosed within the integrated circuit device (101) to implement, using the matrices (207), the computations of the Artificial Neural Network (201 ).
- the integrated circuit device (101) has a first interface (107) configured to be connected to a memory controller (227) and a second interface (221 ) configured to be connected to an image generator, such as a camera (223), radar, lidar, ultrasound scanner, or medical imaging equipment.
- the second interface (221 ) implements a camera command interface protocol, and/or a Mobile Industry Processor Interface (MIPI) protocol.
- MIPI Mobile Industry Processor Interface
- the second interface (221) is configured to be connected to a camera (223) via a serial bus connection (228) in accordance with a Mobile Industry Processor Interface (MIPI) Camera Serial Interface (CSI) protocol.
- MIPI Mobile Industry Processor Interface
- CSI Camera Serial Interface
- the integrated circuit device (101) communicates, via the second interface (221), with the image generator (e.g., 223) to receive image data as input (221) to the Artificial Neural Network (201).
- the image generator e.g., 223
- the integrated circuit device (101) executes the instructions (205) to generate output (213) from the Artificial Neural Network (201) according to the input (211 ).
- the integrated circuit device (101) stores, in random access memory (105) that is enclosed within the integrated circuit device (101), the output (213) from the Artificial Neural Network (201).
- the integrated circuit device (101) provides, via the first interface (107), the output (213) to a Central Processing Unit (225) that is in control of the memory controller (227).
- the integrated circuit device (101) can buffer, into the random access memory (105) via the second interface (221 ) and as input to the Artificial Neural Network (201 ), a next set of image data from the image generator (e.g., 223), while simultaneously providing, from the random access memory (105) via the first interface (107) and concurrently with the buffering, output previously generated from the Artificial Neural Network (201 ) according to a prior set of image data from the image generator (e.g., 223).
- a next set of image data from the image generator e.g., 223
- the integrated circuit device (101) can buffer, into the random access memory (105) via the second interface (221 ) and as input to the Artificial Neural Network (201 ), a next set of image data from the image generator (e.g., 223), while simultaneously providing, from the random access memory (105) via the first interface (107) and concurrently with the buffering, output previously generated from the Artificial Neural Network (201 ) according to a prior set of image data from the image generator (e.g.,
- the integrated circuit device (101) can be enclosed within an integrated circuit package.
- the integrated circuit device (101) has a Deep Learning Accelerator (103) with processing units (111), a control unit (113) and local memory (115).
- the processing units (111) include at least a matrix-matrix unit (121) configured to execute an instruction having two matrix operands; the matrix-matrix unit (121) includes a plurality of matrix-vector units (141 to 143) configured to operate in parallel; each of the matrix-vector units includes a plurality of vector-vector units (161 to 163) configured to operate in parallel; and each of the vector-vector units includes a plurality of multiply-accumulate units (171 to 173) configured to operate in parallel.
- the integrated circuit device (101) can transmit, via the second interface (221), a command to the image generator (e g., 223), where the command instructs the image generator (e.g., 223) to adjust resolution or frame rate of image data generated by the image generator (e.g., 223).
- a command to the image generator (e g., 223), where the command instructs the image generator (e.g., 223) to adjust resolution or frame rate of image data generated by the image generator (e.g., 223).
- the integrated circuit device (101) can transmit, via the second interface (221), another command to the image generator (e.g., 223) to instruct the image generator (e.g., 223) to adjust an area of interest, a point of view, or a field of view of image data generated by the image generator (e.g., 223).
- another command to the image generator (e.g., 223) to instruct the image generator (e.g., 223) to adjust an area of interest, a point of view, or a field of view of image data generated by the image generator (e.g., 223).
- a typical data processing system may include an inter-connect (e.g., bus and system core logic), which interconnects a microprocessor(s) and memory.
- the microprocessor is typically coupled to cache memory.
- the inter-connect interconnects the microprocessor(s) and the memory together and also interconnects them to input/output (I/O) device(s) via I/O controller(s).
- I/O devices may include a display device and/or peripheral devices, such as mice, keyboards, modems, network interfaces, printers, scanners, video cameras and other devices known in the art.
- the data processing system is a server system, some of the I/O devices, such as printers, scanners, mice, and/or keyboards, are optional.
- the inter-connect can include one or more buses connected to one another through various bridges, controllers and/or adapters.
- the I/O controllers include a USB (Universal Serial Bus) adapter for controlling USB peripherals, and/or an IEEE-1394 bus adapter for controlling IEEE-1394 peripherals.
- the memory may include one or more of: ROM (Read Only Memory), volatile RAM (Random Access Memory), and non-volatile memory, such as hard drive, flash memory, etc.
- Volatile RAM is typically implemented as dynamic RAM (DRAM) which requires power continually in order to refresh or maintain the data in the memory.
- DRAM dynamic RAM
- Non-volatile memory is typically a magnetic hard drive, a magnetic optical drive, an optical drive (e.g., a DVD RAM), or other type of memory system which maintains data even after power is removed from the system.
- the non-volatile memory may also be a random access memory.
- the non-volatile memory can be a local device coupled directly to the rest of the components in the data processing system.
- a non-volatile memory that is remote from the system such as a network storage device coupled to the data processing system through a network interface such as a modem or Ethernet interface, can also be used.
- the functions and operations as described here can be implemented using special purpose circuitry, with or without software instructions, such as using Application-Specific Integrated Circuit (ASIC) or Field- Programmable Gate Array (FPGA).
- ASIC Application-Specific Integrated Circuit
- FPGA Field- Programmable Gate Array
- Embodiments can be implemented using hardwired circuitry without software instructions, or in combination with software instructions. Thus, the techniques are limited neither to any specific combination of hardware circuitry and software, nor to any particular source for the instructions executed by the data processing system.
- At least some aspects disclosed can be embodied, at least in part, in software. That is, the techniques may be carried out in a computer system or other data processing system in response to its processor, such as a microprocessor, executing sequences of instructions contained in a memory, such as ROM, volatile RAM, non-volatile memory, cache or a remote storage device.
- processor such as a microprocessor
- a memory such as ROM, volatile RAM, non-volatile memory, cache or a remote storage device.
- Routines executed to implement the embodiments may be implemented as part of an operating system or a specific application, component, program, object, module or sequence of instructions referred to as “computer programs.”
- the computer programs typically include one or more instructions set at various times in various memory and storage devices in a computer, and that, when read and executed by one or more processors in a computer, cause the computer to perform operations necessary to execute elements involving the various aspects.
- a machine readable medium can be used to store software and data which when executed by a data processing system causes the system to perform various methods.
- the executable software and data may be stored in various places including for example ROM, volatile RAM, non-volatile memory and/or cache. Portions of this software and/or data may be stored in any one of these storage devices.
- the data and instructions can be obtained from centralized servers or peer to peer networks. Different portions of the data and instructions can be obtained from different centralized servers and/or peer to peer networks at different times and in different communication sessions or in a same communication session.
- the data and instructions can be obtained in entirety prior to the execution of the applications. Alternatively, portions of the data and instructions can be obtained dynamically, just in time, when needed for execution. Thus, it is not required that the data and instructions be on a machine readable medium in entirety at a particular instance of time.
- Examples of computer-readable media include but are not limited to non- transitory, recordable and non-recordable type media such as volatile and non volatile memory devices, Read Only Memory (ROM), Random Access Memory (RAM), flash memory devices, floppy and other removable disks, magnetic disk storage media, optical storage media (e.g., Compact Disk Read-Only Memory (CD ROM), Digital Versatile Disks (DVDs), etc.), among others.
- the computer-readable media may store the instructions.
- the instructions may also be embodied in digital and analog communication links for electrical, optical, acoustical or other forms of propagated signals, such as carrier waves, infrared signals, digital signals, etc.
- propagated signals such as carrier waves, infrared signals, digital signals, etc. are not tangible machine readable medium and are not configured to store instructions.
- a machine readable medium includes any mechanism that provides (i.e. , stores and/or transmits) information in a form accessible by a machine (e.g., a computer, network device, personal digital assistant, manufacturing tool, any device with a set of one or more processors, etc.).
- hardwired circuitry may be used in combination with software instructions to implement the techniques.
- the techniques are neither limited to any specific combination of hardware circuitry and software nor to any particular source for the instructions executed by the data processing system.
- the above description and drawings are illustrative and are not to be construed as limiting. Numerous specific details are described to provide a thorough understanding. However, in certain instances, well known or conventional details are not described in order to avoid obscuring the description. References to one or an embodiment in the present disclosure are not necessarily references to the same embodiment; and, such references mean at least one.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- Software Systems (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Biomedical Technology (AREA)
- Life Sciences & Earth Sciences (AREA)
- Health & Medical Sciences (AREA)
- Biophysics (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Mathematical Physics (AREA)
- Computing Systems (AREA)
- Multimedia (AREA)
- Signal Processing (AREA)
- Data Mining & Analysis (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Artificial Intelligence (AREA)
- Neurology (AREA)
- Computer Hardware Design (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Advance Control (AREA)
Priority Applications (3)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN202180025761.0A CN115443468A (zh) | 2020-04-09 | 2021-04-06 | 具有摄像头接口的深度学习加速器及随机存取存储器 |
KR1020227038578A KR20220164553A (ko) | 2020-04-09 | 2021-04-06 | 카메라 인터페이스를 구비한 딥 러닝 가속기 및 랜덤 액세스 메모리 |
DE112021002239.3T DE112021002239T5 (de) | 2020-04-09 | 2021-04-06 | Deep learning-beschleuniger und direktzugriffsspeicher mit einer kameraschnittstelle |
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US16/844,997 | 2020-04-09 | ||
US16/844,997 US11355175B2 (en) | 2020-04-09 | 2020-04-09 | Deep learning accelerator and random access memory with a camera interface |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2021207237A1 true WO2021207237A1 (en) | 2021-10-14 |
Family
ID=78006978
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US2021/026008 WO2021207237A1 (en) | 2020-04-09 | 2021-04-06 | Deep learning accelerator and random access memory with a camera interface |
Country Status (5)
Country | Link |
---|---|
US (2) | US11355175B2 (de) |
KR (1) | KR20220164553A (de) |
CN (1) | CN115443468A (de) |
DE (1) | DE112021002239T5 (de) |
WO (1) | WO2021207237A1 (de) |
Families Citing this family (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US11874897B2 (en) | 2020-04-09 | 2024-01-16 | Micron Technology, Inc. | Integrated circuit device with deep learning accelerator and random access memory |
US11461651B2 (en) | 2020-04-09 | 2022-10-04 | Micron Technology, Inc. | System on a chip with deep learning accelerator and random access memory |
US11355175B2 (en) | 2020-04-09 | 2022-06-07 | Micron Technology, Inc. | Deep learning accelerator and random access memory with a camera interface |
US11726784B2 (en) | 2020-04-09 | 2023-08-15 | Micron Technology, Inc. | Patient monitoring using edge servers having deep learning accelerator and random access memory |
US11887647B2 (en) | 2020-04-09 | 2024-01-30 | Micron Technology, Inc. | Deep learning accelerator and random access memory with separate memory access connections |
Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170316312A1 (en) * | 2016-05-02 | 2017-11-02 | Cavium, Inc. | Systems and methods for deep learning processor |
US20180307974A1 (en) * | 2017-04-19 | 2018-10-25 | Beijing Deephi Intelligence Technology Co., Ltd. | Device for implementing artificial neural network with mutiple instruction units |
US20190171941A1 (en) * | 2017-12-01 | 2019-06-06 | Abee Technology Co., Ltd. | Electronic device, accelerator, and accelerating method applicable to convolutional neural network computation |
US20190180170A1 (en) * | 2017-12-12 | 2019-06-13 | Amazon Technologies, Inc. | Multi-memory on-chip computational network |
WO2020058560A1 (en) * | 2018-09-17 | 2020-03-26 | Nokia Solutions And Networks Oy | Object tracking |
Family Cites Families (61)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US5815458A (en) | 1996-09-06 | 1998-09-29 | Micron Technology, Inc. | System and method for writing data to memory cells so as to enable faster reads of the data using dual wordline drivers |
US7251376B2 (en) | 2003-08-29 | 2007-07-31 | Canadian Space Agency | Data compression engines and real-time wideband compressor for multi-dimensional data |
CA2559241C (en) | 2004-03-12 | 2015-10-13 | Olivier Saidi | Systems and methods for treating, diagnosing and predicting the occurrence of a medical condition |
US7873812B1 (en) | 2004-04-05 | 2011-01-18 | Tibet MIMAR | Method and system for efficient matrix multiplication in a SIMD processor architecture |
US20060230213A1 (en) * | 2005-03-29 | 2006-10-12 | Via Technologies, Inc. | Digital signal system with accelerators and method for operating the same |
US10754764B2 (en) | 2018-04-22 | 2020-08-25 | Sas Institute Inc. | Validation sets for machine learning algorithms |
US8655817B2 (en) | 2008-02-20 | 2014-02-18 | Digital Medical Experts Inc. | Expert system for determining patient treatment response |
US8131659B2 (en) | 2008-09-25 | 2012-03-06 | Microsoft Corporation | Field-programmable gate array based accelerator system |
NZ572036A (en) | 2008-10-15 | 2010-03-26 | Nikola Kirilov Kasabov | Data analysis and predictive systems and related methodologies |
US8458377B2 (en) | 2010-03-05 | 2013-06-04 | Lsi Corporation | DMA engine capable of concurrent data manipulation |
EP2601613B1 (de) * | 2011-04-06 | 2020-08-12 | BlackBerry Limited | Effiziente umsetzung von hash-algorithmen auf einem prozessor |
US9199122B2 (en) | 2012-10-09 | 2015-12-01 | Kc Holdings I | Personalized avatar responsive to user physical state and context |
US11503031B1 (en) | 2015-05-29 | 2022-11-15 | Pure Storage, Inc. | Storage array access control from cloud-based user authorization and authentication |
US11126549B2 (en) | 2016-03-31 | 2021-09-21 | Hewlett Packard Enterprise Development Lp | Processing in-memory architectures for performing logical operations |
GB2553783B (en) | 2016-09-13 | 2020-11-04 | Advanced Risc Mach Ltd | Vector multiply-add instruction |
US10127495B1 (en) | 2017-04-14 | 2018-11-13 | Rohan Bopardikar | Reducing the size of a neural network through reduction of the weight matrices |
CN107679621B (zh) | 2017-04-19 | 2020-12-08 | 赛灵思公司 | 人工神经网络处理装置 |
GB2563878B (en) | 2017-06-28 | 2019-11-20 | Advanced Risc Mach Ltd | Register-based matrix multiplication |
US20190027018A1 (en) | 2017-07-21 | 2019-01-24 | Accenture Global Solutions Limited | Artificial intelligence based service control and home monitoring |
US10542889B2 (en) | 2017-08-14 | 2020-01-28 | Amrita Vishwa Vidyapeetham | Systems, methods, and devices for remote health monitoring and management |
US11204747B1 (en) | 2017-10-17 | 2021-12-21 | Xilinx, Inc. | Re-targetable interface for data exchange between heterogeneous systems and accelerator abstraction into software instructions |
US10515135B1 (en) | 2017-10-17 | 2019-12-24 | Xilinx, Inc. | Data format suitable for fast massively parallel general matrix multiplication in a programmable IC |
EP3730040A4 (de) | 2017-12-20 | 2021-10-06 | Medi Whale Inc. | Verfahren und vorrichtung zur unterstützung der diagnose einer kardiovaskulären erkrankung |
US10885314B2 (en) | 2018-01-22 | 2021-01-05 | Kneron Inc. | Face identification system and face identification method with high security level and low power consumption |
US10545559B2 (en) * | 2018-02-20 | 2020-01-28 | Apical Limited | Data processing system and method |
US20190286973A1 (en) | 2018-03-14 | 2019-09-19 | Microsoft Technology Licensing, Llc | Hardware accelerated neural network subgraphs |
US10698730B2 (en) | 2018-04-03 | 2020-06-30 | FuriosaAI Co. | Neural network processor |
KR102589968B1 (ko) | 2018-04-17 | 2023-10-16 | 삼성전자주식회사 | 3 차원 적층 구조를 갖는 뉴로모픽 회로 및 이를 포함하는 반도체 장치 |
US10387122B1 (en) | 2018-05-04 | 2019-08-20 | Olsen Ip Reserve, Llc | Residue number matrix multiplier |
US10754649B2 (en) | 2018-07-24 | 2020-08-25 | Apple Inc. | Computation engine that operates in matrix and vector modes |
WO2020044208A1 (en) * | 2018-08-29 | 2020-03-05 | Cerebras Systems Inc. | Isa enhancements for accelerated deep learning |
US11188825B2 (en) | 2018-10-15 | 2021-11-30 | International Business Machines Corporation | Mixed-precision deep-learning with multi-memristive devices |
US10678479B1 (en) | 2018-11-29 | 2020-06-09 | Amazon Technologies, Inc. | Registers for restricted memory |
US20200183837A1 (en) | 2018-12-07 | 2020-06-11 | Samsung Electronics Co., Ltd. | Dataflow accelerator architecture for general matrix-matrix multiplication and tensor computation in deep learning |
EP3899811A4 (de) | 2018-12-18 | 2022-09-28 | Movidius Ltd. | Kompression eines neuronalen netzes |
US10929503B2 (en) | 2018-12-21 | 2021-02-23 | Intel Corporation | Apparatus and method for a masked multiply instruction to support neural network pruning operations |
US10978382B2 (en) | 2019-01-30 | 2021-04-13 | Taiwan Semiconductor Manufacturing Company, Ltd. | Integrated circuit package and method |
US11961007B2 (en) | 2019-02-06 | 2024-04-16 | Qualcomm Incorporated | Split network acceleration architecture |
US11539763B2 (en) | 2019-02-13 | 2022-12-27 | Apple Inc. | Apparatus and method for session initiated protocol (SIP) registration procedure for access network bitrate recommendation (ANBR) capability signaling |
WO2020172494A1 (en) | 2019-02-22 | 2020-08-27 | Neureality Ltd. | Directed and interconnected grid dataflow architecture |
US11630770B2 (en) | 2019-07-11 | 2023-04-18 | Meta Platforms Technologies, Llc | Systems and methods for reading and writing sparse data in a neural network accelerator |
US20210012178A1 (en) | 2019-07-11 | 2021-01-14 | Facebook Technologies, Llc | Systems, methods, and devices for early-exit from convolution |
US11675998B2 (en) | 2019-07-15 | 2023-06-13 | Meta Platforms Technologies, Llc | System and method for performing small channel count convolutions in energy-efficient input operand stationary accelerator |
EP3933664A4 (de) * | 2019-07-31 | 2022-04-06 | Huawei Technologies Co., Ltd. | Integriertes chip- und sensordatenverarbeitungsverfahren |
US11494608B2 (en) | 2019-08-14 | 2022-11-08 | Intel Corporation | Methods and apparatus to tile walk a tensor for convolution operations |
US11334399B2 (en) | 2019-08-15 | 2022-05-17 | Intel Corporation | Methods and apparatus to manage power of deep learning accelerator systems |
US11500959B2 (en) | 2019-08-16 | 2022-11-15 | Google Llc | Multiple output fusion for operations performed in a multi-dimensional array of processing units |
US11687341B2 (en) | 2019-08-29 | 2023-06-27 | Intel Corporation | Multi-variate strided read operations for accessing matrix operands |
US11366979B2 (en) | 2019-11-14 | 2022-06-21 | Alibaba Group Holding Limited | Using selected components of frequency domain image data in artificial intelligence tasks |
TWI728556B (zh) | 2019-11-18 | 2021-05-21 | 財團法人工業技術研究院 | 神經元電路及類神經網路晶片 |
US20210192287A1 (en) | 2019-12-18 | 2021-06-24 | Nvidia Corporation | Master transform architecture for deep learning |
CN111240743B (zh) | 2020-01-03 | 2022-06-03 | 格兰菲智能科技有限公司 | 人工智能集成电路 |
US20210279629A1 (en) | 2020-03-05 | 2021-09-09 | oneFiveFIFTY LLC | Machine learning and computer-based generation of standard work matrices for improving execution of a standard work |
US11874897B2 (en) | 2020-04-09 | 2024-01-16 | Micron Technology, Inc. | Integrated circuit device with deep learning accelerator and random access memory |
US20210320967A1 (en) | 2020-04-09 | 2021-10-14 | Micron Technology, Inc. | Edge Server with Deep Learning Accelerator and Random Access Memory |
US11887647B2 (en) | 2020-04-09 | 2024-01-30 | Micron Technology, Inc. | Deep learning accelerator and random access memory with separate memory access connections |
US11726784B2 (en) | 2020-04-09 | 2023-08-15 | Micron Technology, Inc. | Patient monitoring using edge servers having deep learning accelerator and random access memory |
US11461651B2 (en) | 2020-04-09 | 2022-10-04 | Micron Technology, Inc. | System on a chip with deep learning accelerator and random access memory |
US11355175B2 (en) | 2020-04-09 | 2022-06-07 | Micron Technology, Inc. | Deep learning accelerator and random access memory with a camera interface |
US11500811B2 (en) | 2020-06-12 | 2022-11-15 | Alibaba Group Holding Limited | Apparatuses and methods for map reduce |
US11544548B2 (en) | 2021-05-24 | 2023-01-03 | Rebellions Inc. | Processing element and neural processing device including same |
-
2020
- 2020-04-09 US US16/844,997 patent/US11355175B2/en active Active
-
2021
- 2021-04-06 WO PCT/US2021/026008 patent/WO2021207237A1/en active Application Filing
- 2021-04-06 CN CN202180025761.0A patent/CN115443468A/zh active Pending
- 2021-04-06 KR KR1020227038578A patent/KR20220164553A/ko unknown
- 2021-04-06 DE DE112021002239.3T patent/DE112021002239T5/de active Pending
-
2022
- 2022-04-26 US US17/729,830 patent/US11942135B2/en active Active
Patent Citations (5)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20170316312A1 (en) * | 2016-05-02 | 2017-11-02 | Cavium, Inc. | Systems and methods for deep learning processor |
US20180307974A1 (en) * | 2017-04-19 | 2018-10-25 | Beijing Deephi Intelligence Technology Co., Ltd. | Device for implementing artificial neural network with mutiple instruction units |
US20190171941A1 (en) * | 2017-12-01 | 2019-06-06 | Abee Technology Co., Ltd. | Electronic device, accelerator, and accelerating method applicable to convolutional neural network computation |
US20190180170A1 (en) * | 2017-12-12 | 2019-06-13 | Amazon Technologies, Inc. | Multi-memory on-chip computational network |
WO2020058560A1 (en) * | 2018-09-17 | 2020-03-26 | Nokia Solutions And Networks Oy | Object tracking |
Also Published As
Publication number | Publication date |
---|---|
DE112021002239T5 (de) | 2023-02-09 |
US11942135B2 (en) | 2024-03-26 |
US11355175B2 (en) | 2022-06-07 |
US20210319823A1 (en) | 2021-10-14 |
KR20220164553A (ko) | 2022-12-13 |
US20220254400A1 (en) | 2022-08-11 |
CN115443468A (zh) | 2022-12-06 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US11942135B2 (en) | Deep learning accelerator and random access memory with a camera interface | |
US11461651B2 (en) | System on a chip with deep learning accelerator and random access memory | |
US20210320967A1 (en) | Edge Server with Deep Learning Accelerator and Random Access Memory | |
US11874897B2 (en) | Integrated circuit device with deep learning accelerator and random access memory | |
US11887647B2 (en) | Deep learning accelerator and random access memory with separate memory access connections | |
US20220188606A1 (en) | Memory Configuration to Support Deep Learning Accelerator in an Integrated Circuit Device | |
WO2022031446A1 (en) | Optimized sensor fusion in deep learning accelerator with integrated random access memory | |
US20220043502A1 (en) | Intelligent low power modes for deep learning accelerator and random access memory | |
US20230161936A1 (en) | Integrated Sensor Device with Deep Learning Accelerator and Random Access Memory | |
US20220044101A1 (en) | Collaborative sensor data processing by deep learning accelerators with integrated random access memory | |
US20220147808A1 (en) | Compiler configurable to generate instructions executable by different deep learning accelerators from a description of an artificial neural network | |
US20220147809A1 (en) | Deep learning accelerators with configurable hardware options optimizable via compiler | |
US20220147811A1 (en) | Implement the computation of an artificial neural network using multiple deep learning accelerators | |
US20220043696A1 (en) | Distributed inferencing using deep learning accelerators with integrated random access memory |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 21783865 Country of ref document: EP Kind code of ref document: A1 |
|
ENP | Entry into the national phase |
Ref document number: 20227038578 Country of ref document: KR Kind code of ref document: A |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 21783865 Country of ref document: EP Kind code of ref document: A1 |